how ever, only be regarded as a guideline for a possible size effect.52Note that the associated R2is relatively low (0.119) and therefore one has to exercise caution in interpreting the results. The first step one can take is to test for robustness of the size effect by running multiple regressions, where some important control variables are introduced (in Section 3.2.4). 3.2.3 Multiple regressions as the proper empirical tool To address the relationship between government size and country size more comprehensively, it is necessary to include several variables – like, for instance, geographic dummies – which have been considered to be worth controlling for in Section 3.1 from a theoretical viewpoint. A mul- tiple regression analysis is an appropriate tool to analyze the question at hand, because it allows us to obtain a quantitative assessment of the rela - tion ship between country size and government size without neglecting other important determinants of government size (like per capita GDP, dummy variables for groups of countries, population density, etc.). It has to be mentioned here that the analysis is designed to shed light on the overall picture of the issue and that it can only investigate the cost (supply) side of the provision of public goods. The possible and expected result that small countries have to bear higher costs for the pro- vision of public goods cannot be the sole argument to declare them eco- nomically inferior (from an efficiency perspective) to larger countries. The results of this chapter have to be supplemented by a thorough inves - tigation of the demand side effects of smallness (preference adequacy), other, hardly-quantifiable effects of smallness («distance» to politicians and bureaucrats, advantages/disadvantages of sovereignty, advantages/ disadvantages of homogeneity, etc.) and the sources of the possible cost disadvantage (diseconomies of scale, lack of competition, etc.). More - over, there are, of course, other sources of inefficiency prevalent in any country which are much easier to adjust than country size, if the latter is adjustable at all. Most of the important determinants, especially on the 53 
Empirical evidence 52Nevertheless, it can be shown by sensitivity analyses that the general picture does not change if UNDP data its used instead of IMF data for the 90ies.